What B2B SaaS becomes after AI

There is a confident story going around that AI is going to eat B2B SaaS. The logic is simple enough. If a model can write software, then any company can generate the tool it needs instead of paying a vendor for it. Why buy a ticketing system when you can produce one on a Tuesday afternoon.

I do not think that is quite right, but I do not think it is entirely wrong either. Having spent the last while building software inside a company rather than selling it, my honest read is that AI is not killing B2B SaaS so much as moving the value somewhere else. The parts that were expensive are getting cheap. The parts that were taken for granted are becoming the whole game.

What actually got cheap

The thing AI made cheap is the building of features. Screens, forms, workflows, reports, the CRUD layer that most business software is made of. That used to take a team and a roadmap. Now it takes far less.

Which means the moat around "we have a tool that does X" has mostly evaporated. If your product's defensibility was that you had built a competent interface on top of a database, you are in trouble, because competent interfaces are now a commodity. That is the grain of truth in the AI-eats-SaaS story.

What did not get cheap

Here is the part the story skips. Software inside a business is never just the interface. It sits in the middle of a mess of other things, and those things did not get cheaper at all.

The data, and agreeing what it means. Every company I have worked in has argued about what an "active customer" is or when revenue counts. AI does not settle that argument. It just makes it faster to build something on top of whichever answer you pick, which honestly makes the argument more urgent, not less.

Integrations, and the boring plumbing. Real systems have to talk to the billing system, the CRM, the warehouse, the compliance rules. That is grinding, unglamorous work that has to keep working when someone else changes their API.

Trust and adoption. This is the one I keep coming back to. Getting people to actually use a tool, and to believe it enough to act on what it says, has never been a technology problem. It is a human one. Cheaper software does not make adoption easier. It makes it harder, because now there is more software competing for the same limited attention.

Accountability. When something goes wrong, somebody has to be responsible. A vendor with a contract can be. A tool you generated last quarter and nobody owns cannot.

So where does the value go

My guess is that it moves in two directions at once.

Downward, into the boring infrastructure. The systems of record, the data layers, the integration plumbing, the compliance and audit trails. Unsexy, hard to replicate, and now more valuable relative to everything sitting on top of them.

And upward, into judgment. Not "here is a screen showing your data" but "here is what is happening, here is what it probably means, and here is the decision in front of you." That layer is hard to fake, because it requires actually understanding the business, not just rendering its data.

The squeezed middle is generic feature software. Products whose pitch was a nice interface over a database, with no deep data, no real integrations, and no judgment on top. That middle was always the most crowded part of the market. It is now the most exposed.

What this means for how you build

A few things follow, and they are more about discipline than technology.

Stop treating features as the product. If a feature can be generated in an afternoon, it is not your differentiation, no matter how much of the roadmap it occupied.

Get obsessive about the data underneath. The agreed definitions, the quality, the coverage. That is the part that compounds and the part a competitor cannot summon out of thin air.

Design for trust as a first-class feature. In a world where software is abundant and often unreliable, the product people keep using will be the one that shows its work and is honest about what it does not know. Abundance makes trust scarcer, not less important.

Solve the workflow, not the screen. The value is in fitting into how work actually happens, across the messy handoffs between people and systems. That is where the real difficulty lives, and it is not something a model generates for you.

The unglamorous conclusion

I think we end up with more software, not less, and a much lower tolerance for software that does not earn its place. Buyers will keep paying for things that are genuinely hard: deep data, real integration, accountability, and judgment they can trust. They will stop paying for things they can now make themselves.

That is less dramatic than "AI kills SaaS," but I think it is closer to what is happening. The interface stopped being the moat. What sits underneath it, and what sits above it, became the whole business.